Similarity-driven chunking during encoding supports biased but more precise working memory

Visual working memory (VWM) allows us to maintain and manipulate information in service of behavioral goals. Navigating rich visual environments often involves holding multiple items in VWM—some of them very similar. Recent work suggests that inter-item similarity impairs memory precision during encoding but enhances precision during active memory maintenance. The present study tested whether this inter-item similarity benefit observed during memory maintenance was due to chunking similar items into summary representations. In Experiment 1, participants encoded sample displays with four colored circles into memory: two circles were similar to each other in color, two items were dissimilar to both each other and to the similar items. Using a retrospective cue (“retro-cue”) presented after the sample display, we manipulated the inter-item similarity of the remembered stimuli by cueing two similar items, one similar and one dissimilar item, or two dissimilar items. Consistent with prior work, we observed higher precision memory and attractive biases between similar items, consistent with chunking. In Experiment 2 we observed a similarity benefit and attractive bias for similar items even in the absence of a retro-cue. Importantly, the magnitude of the similarity benefit and attractive bias was the same across valid and neutral cues, suggesting that similarity-based chunking occurs early during encoding. In Experiment 3, we manipulated the onset of the retro-cue to occur early during the delay period, and we replicated the results of Experiments 1 and 2. Together, these results suggest that inter-item similarity enhances VWM performance through chunking that occurs early during encoding as opposed to during maintenance in memory.